Interpretable machine-learning enhanced parametrization methodology for Pluronics-water mixtures in DPD simulations.

Area of Science:

• Computational chemistry
• Materials science
• Machine learning applications

Background:

• Dissipative Particle Dynamics (DPD) is vital for simulating structured fluids but faces parameterization challenges.
• Accurate physical property replication requires precise model parameters, which are difficult to determine.
• High computational costs of DPD simulations hinder extensive exploration and optimization.

Purpose of the Study:

• To integrate machine learning into the parameterization of Pluronic systems using DPD.
• To develop a data-driven workflow for accurate model parameter determination.
• To enhance the efficiency and interpretability of Pluronic system calibration.

Main Methods:

• Utilized Gaussian Process Regression (GPR) for building surrogate models of DPD simulations.
• Employed SHAP (SHapley Additive Explanations) analysis for model interpretability.
• Developed a workflow combining GPR and SHAP for efficient parameter optimization.

Main Results:

• GPR-based surrogate models accurately replicated DPD simulation outcomes.
• The integrated approach significantly reduced computational cost and simulation time.
• SHAP analysis provided insights into parameter-property relationships and causal mechanisms.

Conclusions:

• The combined GPR and SHAP approach offers an interpretable machine learning solution for DPD parameterization.
• This methodology streamlines the optimization process for Pluronic systems.
• The work provides a foundation for generalizing parameterization across various conditions and Pluronic systems.